TL;DR
Researchers have developed a contact tracing system that uses "Crypto-spatiotemporal Information" (CSI) to track COVID-19 exposure without sacrificing individual privacy. By combining blockchain for verifiable data storage and Intel SGX (a Trusted Execution Environment) for secure data matching, the system allows for high-precision tracing while keeping raw location data invisible even to system administrators.
What Problem Does This Paper Address?
Effective contact tracing is a cornerstone of pandemic management, but it presents a classic "privacy vs. utility" trade-off.
Existing solutions generally fall into two camps:
- Centralized Systems: (e.g., China’s Health Code) These are highly effective but require users to surrender sensitive location data to a central authority, raising significant surveillance concerns.
- Decentralized Bluetooth Systems: (e.g., Google/Apple Exposure Notifications) These protect privacy better but can be less accurate due to signal interference and lack the "verifiability" of actual location data.
The authors address the gap for a system that uses high-fidelity spatiotemporal (location + time) data—which is more accurate than Bluetooth proximity—without creating a "Big Brother" database of every citizen's movements.
Methods
The authors propose a three-pillar architecture:
- Crypto-spatiotemporal Information (CSI): Location data is converted into a Geohash (a string representing a geographic bounding box) and paired with a Unix timestamp. This data is then encrypted using AES-GCM.
- Blockchain (Hyperledger Fabric): The encrypted CSI is uploaded to a blockchain. This ensures that the records are unalterable and verifiable. The blockchain acts as a decentralized storage layer where users only submit transaction IDs to the inquiry server rather than raw data.
- Trusted Execution Environment (TEE): This is the "secret sauce." The system uses Intel SGX to create a secure "enclave"—an isolated area of a computer's memory. The decryption keys and the matching algorithms live only inside this enclave. When a user wants to check for exposure, the encrypted data is sent into the enclave, decrypted, compared against known positive cases, and the result is sent back. Crucially, the raw data is never visible to the host operating system or the server administrator.
Datasets
The paper does not use a standard public dataset. Instead, the authors evaluated their system through:
- Small-scale Field Experiment: Conducted at Waseda University, involving both indoor (using facial recognition and Wi-Fi locators) and outdoor (using GNSS/GPS) scenarios to generate real-world CSI.
- Simulated Large-scale Data: To test scalability, they simulated data loads based on the population of Japan (approx. 126 million people) to estimate transaction throughput and memory requirements.
Key Insights and Ideas
- The "Black Box" Matching: The use of TEEs solves the fundamental privacy hurdle. It allows a server to "know" if two people were in the same place at the same time without the server actually "seeing" where that place was.
- Geohash Efficiency: By using Geohash strings, the system can quickly judge proximity. Since Geohashes are hierarchical, a shared prefix between two strings indicates spatial closeness, making the matching algorithm computationally efficient.
- Verifiability via Third Parties: To prevent users from "faking" their location (either to hide or to falsely claim exposure), the system incorporates "third-party locators" like Wi-Fi access points and base stations to verify the user's current spatiotemporal info before it is committed to the blockchain.
Key Results
- Data Efficiency: A single CSI entry is tiny—only about 0.104 KB. Even with a high logging frequency (every 5 seconds), a user only generates about 1.8 MB of data per day.
- Scalability: The simulation suggests that for a country like Japan, the system would need at least seven parallel blockchain chains to handle the estimated 58 million daily transactions.
- Performance: In the TEE, matching a user's 14-day history (approx. 40,000 CSI entries) against a database of positive cases took less than 5 seconds. Even with massive server-side datasets (100 MB of patient data), the judgment time remained around 10 seconds.
Limitations and Future Work
- Hardware Bottlenecks: The performance of the system is heavily dependent on Intel SGX memory. As the volume of positive patient data grows, the enclave initialization time increases significantly.
- Adoption Hurdles: Like all contact tracing apps, the system's effectiveness depends on widespread adoption. The authors note that "insufficient users can affect the efficiency of the system."
- Experimental Scale: While the field experiment proved the concept, it was limited in size. Future work would need to address the complexities of a city-wide deployment, including network latency and diverse hardware compatibility.
- Battery Consumption: Continuous GNSS logging and encryption on mobile devices can be power-intensive, a factor not deeply explored in this paper.
Summary updated